Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75664.18592059804,
-      "required_pages_per_second_upper_estimate": 0.006558379866471461
+      "required_token_positions_per_second": 74033.11514398507,
+      "required_pages_per_second_upper_estimate": 0.006417002785465737
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9458.023240074755,
-      "required_pages_per_second_upper_estimate": 0.0037729468805149016
+      "required_token_positions_per_second": 9254.139392998133,
+      "required_pages_per_second_upper_estimate": 0.003691614565580873
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18916.04648014951,
-      "required_pages_per_second_upper_estimate": 0.0043758944421348625
+      "required_token_positions_per_second": 18508.278785996266,
+      "required_pages_per_second_upper_estimate": 0.00428156456255882
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28374.06972022427,
-      "required_pages_per_second_upper_estimate": 0.0017278319787229024
+      "required_token_positions_per_second": 27762.418178994398,
+      "required_pages_per_second_upper_estimate": 0.0016905856089496287
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28374.06972022427,
-      "required_pages_per_second_upper_estimate": 0.0021843144315149185
+      "required_token_positions_per_second": 27762.418178994398,
+      "required_pages_per_second_upper_estimate": 0.00213722780271121
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28374.06972022427,
-      "required_pages_per_second_upper_estimate": 0.352310984022552
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27762.418178994398,
+      "required_pages_per_second_upper_estimate": 0.34471631894650157
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08120945194534102,
-    "producer_p95_seconds": 0.28856329299742356,
-    "gpu_update_seconds": 0.9744108008691208,
-    "producer_batches_per_second": 12.313837565029521,
-    "gpu_batches_per_second": 1.0262612022650561,
-    "producer_headroom": 11.99873632351268,
+    "producer_batch_seconds": 0.5529561846170736,
+    "producer_p95_seconds": 0.352976365997165,
+    "gpu_update_seconds": 0.9958786666832585,
+    "producer_batches_per_second": 1.8084615523244534,
+    "gpu_batches_per_second": 1.0041383889971933,
+    "producer_headroom": 1.8010082794768125,
     "sustainable": true,
-    "buffer_seconds": 60.41346965388549,
+    "buffer_seconds": 60.74859866767877,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 417080.7090499999,
-    "upload_bytes_per_second": 5690581.987380471,
+    "generation_bytes_per_second": 359561.20330583,
+    "upload_bytes_per_second": 5770466.541366329,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.49581102013586,
+    "estimated_full_shards_per_hour": 77.15346407300163,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.2951343000004272,
-      "bytes": 7367275,
-      "pages": 27,
-      "shards": 3
-    },
     {
       "seconds": 1.1229757969995262,
       "bytes": 5828242,
@@ -562,6 +556,12 @@
       "bytes": 882161,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.030573132000427,
+      "bytes": 11789498,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -230,3 +230,4 @@
 {"step": 23000, "loss": 0.6521362792700529, "lr": 0.0003, "grad_norm": 0.13819179859861555, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3365135354, "tokens_per_second": 149722.4295056493, "data_wait_s": 1.4397999620996416e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 23100, "loss": 0.6185001358389854, "lr": 0.0003, "grad_norm": 0.1349197055425745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3379871034, "tokens_per_second": 149380.8185538223, "data_wait_s": 1.4163997548166662e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 23200, "loss": 0.6290844772011042, "lr": 0.0003, "grad_norm": 0.1338471451423039, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3394519508, "tokens_per_second": 148713.17656576616, "data_wait_s": 1.089999932446517e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 23300, "loss": 0.6249117664992809, "lr": 0.0003, "grad_norm": 0.13694267293745865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3409082308, "tokens_per_second": 131761.32919313002, "data_wait_s": 1.4543002180289477e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:736424025d73ce07b277874ffb2ae52075c69d4959b264fa0a3e8ae5d6fbfa95
-size 488899
+oid sha256:e40d588838b73710b57091331e7eefe615e20ace508e21abe2d34d64de416fd3
+size 491017